On the economic consequences of civil war

By Paul Collier
Centre for the Study of African Economies, University of Oxford,
St Cross Building, Manor Road, Oxford, OX1 3UL; and World Bank
e-mail: paul.collier@economics.oxford.ac.uk and pcollier@worldbank.org
A model of the economic effects of civil war and the post-war period is developed. A
key feature is the adjustment of the capital stock through capital ight. Post-war this
ight can either be reversed or continue, depending partly upon how far the capital
stock has adjusted to the war. The model is tested on data for all civil wars since 1960.
After long civil wars the economy recovers rapidly, whereas after short wars it con-
tinues to decline. We then consider the effect on the composition of economic activity,
distinguishing between war-vulnerable and war-safe activities. Evidence for Uganda
shows such compositional effects to be substantial.
1. Introduction
This paper investigates the consequences of civil war for GDP and its composition.
It focuses particularly upon the behaviour of the economy in the early years of a
peace settlement. Civil wars are liable to be more damaging than international wars
in several respects. They are inevitably fought entirely on the territory of the
country. They are likely to undermine the state: both its institutions such as
property rights, and its organisations such as the police. By contrast, as Herbst
(1991) has argued, international wars tend to strengthen the state. The destruction
wrought by warfare, and the erosion of institutions and organisations, constitute a
deterioration in the economic environment. It might be expected that the ending of
civil wars would produce a consequently large peace dividend. This paper argues
that there is no presumption of a peace dividend from the ending of a civil war.
Specically, whether there is a peace dividend is shown to be contingent upon the
duration of the war. The post-war growth rate can be either well above or well
below that which would have occurred without the war. The end of civil wars can
thus give rise to either a peace dividend or a war overhang effect.
Section 2 sets out the theoretical framework. The argument is that civil war
reduces the desired stock of factors of production. Some factors are better able
than others to leave the country and this gives rise to a gradual exodus of such
factors. The behaviour of GDP during civil war thus reects a gradual adjustment
to a sudden deterioration in the environment. Both during the war and after it, the
effects are consequently on the growth rate of GDP rather than simply on its level.
Section 3 tests the theory on a comprehensive data set including all of the civil wars
which took place during 196092, quantifying the effect of war and its aftermath on
# Oxford University Press 1999 Oxford Economic Papers 51 (1999), 168183 168
growth. Section 4 extends the analysis to the composition of GDP. Section 5
concludes.
2. The effects of civil war on GDP: theory
Although the focus of the paper is on the recovery from civil war, it is useful rst to
consider the processes by which the economy is damaged during the war. The most
obvious way in which civil war damages the economy is through the destruction of
some resources. For example, part of the labour force is killed or maimed and
bridges are blown up. However, civil wars are usually fought with much lower
technology than international wars and so tend to be much less destructive, at least
of physical capital. A second effect is the disruption caused by warfare and the often
concomitant social disorder. For example, some roads become unsafe and so extra
costs are incurred in achieving the same outcome. Civil liberties may be suppressed,
and there is evidence that the suppression of such liberties will tend to reduce the
efciency of public expenditure (Isham et al., 1996). While this is not unique to
civil war, the breakdown of social order and the absence of a clear front line are
more common to civil war than to international war. A third effect is the diversion
of public expenditure from output-enhancing activities. For example, as the army
and its powers are expanded, the police force and the rule of law diminish. The
enforcement costs of contracts consequently rise and the security of property rights
is reduced. The costs of expenditure diversion arising from war have been quanti-
ed by Knight et al. (1996). Fourthly, to the extent that these income losses are
regarded as temporary, there will be dissaving, an effect analytically similar to the
destruction of the capital stock. Finally, in response to the deterioration in the
economic environment, private agents will engage in portfolio substitution, shifting
their assets out of the country. Here, assets should be understood to include human
as well as physical and nancial capital.
Following the restoration of peace the destruction, disruption, diversion, and
dissaving effects are all directly ameliorated. However, the portfolio substitution
effect depends upon private asset choices. The main point of the paper is to draw
the implications of the portfolio substitution effect for post-war economic per-
formance. Although it might seem esoteric to focus upon portfolio choice in the
context of civil war, the sheer scale of capital ight from capital-hostile environ-
ments suggests that its effects on economic performance are likely to be large. For
example, a recent study of how African-owned wealth is divided between domestic
capital and foreign assets nds that the latter constitute 39% of the total (Collier
and Gunning, 1999). Despite foreign exchange controls, which until very recently
have been prevalent in most developing countries, capital has evidently been expa-
triated. All types of domestic capital can gradually be transformed into nancial
capital by reducing gross investment below depreciation rates. With respect to
physical capital, maintenance expenditures can be reduced. With respect to
human capital, households can reduce expenditures on education and training
and send more educated household members abroad. With respect to social capital
paul collier 169
such as reputation, rms can behave opportunistically, depleting reputation in
return for short term prots. The resulting nancial capital can then be removed
from the country through illicit trade (under-invoiced exports and over-invoiced
imports). In addition to these indirect processes of expatriation, some types of
capital can be relocated abroad. Transport equipment and livestock are intrinsically
mobile and so relatively easy to remove from the country. For example, in Mozam-
bique during the 1980s the cattle stock declined by over 80% (Bruck, 1996). In
extremis even less obviously mobile physical capital can be removed. Kapuscinski
(1988) describes the astonishing range of capital goods which were packed into
crates and dispatched abroad at the onset of the Angolan civil war. A further
instance of capital expatriation is the emigration of those with skills.
However, the exodus of factor endowments is a gradual process and the ease of
expatriation differs between types of factor. Whereas some types of capital are
highly endogenous, the supplies of buildings, land, and unskilled labour are
likely to be less responsive.
1
As a stylised representation of these differences, dichot-
omise factor endowments into those which are endogenous with respect to war,
denoted by N, and those which are exogenous, denoted by X.
For concreteness, it is useful to characterise pre-war aggregate output in the
simple CobbDouglas form as
Q a:X
b
N
1b
1
The rate of return on the endogenous factor will be its marginal product minus its
rate of depreciation, d. An initial equilibrium is assumed in which the rate of return
equals the return, r, obtainable on foreign assets, so that
1 bN
b
:aX
b
d r 2
Now consider the effect of civil war on the supply of the endogenous asset. Civil
war reduces the rate of return on N relative to that on the foreign asset. The
disruption and diversion effects both reduce factor productivity. This lowers the
multiplicative constant in (2) from a to a
w
, thereby lowering the rate of return on
N. The destruction effect is equivalent to an increase in the rate of depreciation.
Similarly, the weakening of property rights is equivalent from a private perspective
to a further increase in the rate of depreciation. Denoting the private rate of
depreciation during wartime as d
w
> d, the increase in depreciation again reduces
the rate of return on N.
Assets held abroad will usually be liquid since this characterises most asset
markets in developed countries. Civil war creates temporary uncertainty with
respect to the future return on assets held within the country. If the war persists
the return is lower than if the war ends for good. Such temporary uncertainty
creates a premium upon holding assets in liquid form because such assets now
170 economic consequences of civil war
..........................................................................................................................................................................
1
There are some notable exceptions. For example, the land mines so common in Angola act in part as
land-reducing technical regress, while the mass emigration of Hutus from Rwanda substantially reduced
the labour force.
have an option value in addition to their underlying rate of return, raising the
overall return to r
w
. The resulting increased attractiveness of liquid assets during
civil war is an instance of the `bad news principle' (Dixit and Pindyck, 1994).
As can be seen from (2), by increasing d and r while lowering a, civil war
encourages portfolio diversion from N into foreign assets. The desired stock of
endogenous factors will fall. This gives rise to a phase during which the actual stock
of the variable factor is adjusting to its new desired level. The implication for GDP
is that civil war would reduce output both through the disruption and diversion
effects which are one-off because they directly reduce income, and through a phase
of reduced growth as the stock of N declines relative to its counterfactual through
destruction, dissavings, and capital ight.
Now suppose there is a peace settlement. Although peace reduces the costs of
economic activity in various ways, it does not reduce them to the pre-war level.
With peace the destruction effect ceases, reducing the depreciation rate on capital
back to its pre-war level. The disruption effect may also cease so that there is a one-
off increase in productivity, although some aspects of the effect, such as the sup-
pression of civil liberties, may only recover gradually. The diversion effect will be
reversed by the extent of the scal peace dividend. However, this may be quite
limited so that there can only be a partial restoration of productive public expen-
ditures. Military expenditure might be slow to decline: demobilisation is commonly
delayed through fears of its consequences,
2
and the government army may be
expanded due to the inux of rebel forces, as in Zimbabwe. Even where military
expenditure is reduced there might be little scope for increases in productive public
expenditures: in the closing stages of war governments often adopt scally unsus-
tainable positions. For example, in the nal year of the Ethiopian civil war the scal
decit increased by 8% of GDP.
Because of the persistence of the diversion effect, were the stock of N-capital fully
restored its productivity would be below its pre-war level. This is one reason why
although peace increases the optimal stock of the endogenous factor, it does not
restore it to its counterfactual level. A second reason is that the level of risk is likely
to have increased relative to the pre-war situation: a peace settlement will generally
not restore expectations of peace to their pre-war level. The very fact of the civil war
constitutes news that the society is prone to civil war. Further, it is likely to have
polarised the society, so making it easier to coordinate future rebellions. These
effects may be offset by the peace settlement: the rebellion may have been decisively
vanquished, or peace may be guaranteed by external powers. However, as a general
proposition, societies in the early years after a civil war are probably seen as facing
higher risks than societies which have had long histories of peace. The probability
of future peace thus rises on the news of current peace but the risk of `bad news'
does not fully revert to its pre-war level. This both maintains some, albeit reduced,
option value on liquidity, and depresses the return on N relative to pre-war con-
paul collier 171
..........................................................................................................................................................................
2
See Colletta et al. (1996) and Collier (1994) for a discussion of demobilisation experiences.
ditions. The increased probability of war relative to the pre-war situation reduces
the expected productivity of only partially reversible investment relative to its pre-
war level. Similarly, it raises the expected depreciation rate. Thus, the persistence of
the diversion effect and the only partial credibility of peace both lower the return
on N relative to its pre-war level. While the optimal stock of N increases as a result
of peace, it does not fully recover its counterfactual value.
The response of the economy to peace depends crucially upon whether the stock
of N has already adjusted below its newly optimal level. If it has not then the decline
in N continues as if there had been no peace settlement. This might be termed a war
overhang effect. Conversely, if N has already adjusted below its newly optimal level
then the adjustment process goes into reverse. The supply of the endogenous factor
can be increased by repatriation, giving rise to a gradual peace dividend.
Whether the stock of N has adjusted below its newly optimal level depends upon
many dimensions of the adjustment process, most of which are unlikely to be
measurable. However, one evident and observable dimension is its duration. Ceteris
paribus, the longer the duration of the war the more likely it is that the supply of
the endogenous factor has had the time to adjust below its post-war optimal level.
Hence, the more likely it is that peace will produce a peace dividend rather than a
war overhang.
The foregoing implies four testable hypotheses. First, because civil wars gradually
reduce the stock of the endogenous factor, they reduce the growth rate of GDP
rather than just its level. Secondly, because adjustment of the endogenous factor
may have been incomplete by the time the war ends, peace may not end decline.
Thirdly, because peace may reverse the exodus of the endogenous factor there is the
potential for accelerated growth: that is, there is a potential peace dividend.
Fourthly, the longer has been the war the more likely is there to be a peace dividend
rather than a war overhang. In the limiting case, a momentary war will lower the
desired post-war stock of the endogenous factor without providing any oppor-
tunity for this reduction to be affected, so that the early post-war period will be
characterised by a reduction in the growth rate. That is, sufciently short wars will
be characterised not by a peace dividend but by a war overhang.
3. The effects of civil war on GDP: evidence
In this section the above hypotheses are tested on a comprehensive data set. The
main issue is whether, as implied by the portfolio substitution effect, there is some
critical duration of civil wars below which growth continues to be retarded in the
post-war period, and above which there is a peace dividend.
Although no previous study has quantied the growth effects of civil wars and
their aftermath, two recent papers are highly pertinent. Knight et al. (1996) have
investigated the effects of war and quantied a peace dividend using a 79-country
data set. Two of their results are important for the present study. The rst is their
panel regression of the ratio of investment to GDP. Warfare is found to have a
strongly negative effect, being the most signicant variable in the regression. While
172 economic consequences of civil war
unsurprising, this is consistent with the underlying presumption of Section 2,
namely that civil wars reduce growth mainly by depleting the domestic capital
stock in its various forms. Indeed, when Knight et al. control for physical and
human capital, together with military spending and trade policy, they fail to
nd a signicant effect of war in their growth regression. The second result of
importance concerns their estimate of the peace dividend. The source of the
dividend in their study is entirely a reduction in military expenditure. That is, it
is the type of dividend arising from the end of the Cold War rather than from
the ending of a civil war. Indeed, Knight et al. specically control for warfare in
quantifying the effect of reduced military spending on growth: for a given
propensity to warfare, they measure the gains from reducing the military budget.
Further, they assume that the gains from reducing military expenditure are entirely
symmetrical with the costs from increasing it. Their concept of the peace dividend
from reducing military expenditure is thus close to the pure diversion effect
postulated in Section 2: the cost inicted by the squeeze of military expenditure
on productive public expenditures. The other four effects of a civil war arise from
directly or indirectly from violence rather than from the composition of public
expenditure. Their principal nding, that increased military expenditure has
signicant negative growth effects, thus supports the postulate that the diversion
effect is costly to growth.
The second pertinent study is that of de Melo et al. (1996) which investigates the
effect of civil wars in the transition economies of Eastern Europe on the average
growth rate over the period 198994. They include in their growth regression a
dummy variable `regional tension' for `persistent internal conicts or conict-
related blockades'. The variable is highly signicant and reduces the annual average
growth rate during the ve years by 9%. Since some of the conicts ended before
1994 this conates growth performance during the war with that after the war.
However, because of the limited number of observations no distinction could be
made.
The present paper explicitly quanties the effects of civil war on growth both
during the war and during the rst ve post-war years. The data set used for growth
rates is the now-standard Penn World Tables, covering the period 196089. The
innovation is to combine this with the standard source on civil wars (Small and
Singer, 1982, 1994). This data set gives the dates of the starting and ending of all
civil wars since 1816 by month, dened on a common set of objective criteria.
3
This
permits a focus upon civil war as opposed to the wider concept deployed in both
paul collier 173
..........................................................................................................................................................................
3
Singer and Small dene civil war on three criteria: military action internal to the metropole, active
participation of the national government, and effective resistance by both sides. This denition produces
a classication which is somewhat too restrictive for our purposes. For example, Ethiopia is not classied
as having experienced civil war on these criteria. In the present study their denition has been broadened
slightly to include what they term `extra-systemic' wars, which brings Ethiopia into the set, while still
excluding inter-state wars. A fuller discussion of the Singer and Small data set is given in Collier and
Hoefer (1998).
Knight et al. and the pioneering applied growth studies of Barro (1991). Combin-
ing the data sets provides a sample of 92 countries of which 19 had civil wars.
The dependent variable was the decade average per capita GDP growth rate of
each country between 1960 and 1989. This makes the study directly comparable
with other recent work on the determinants of growth such as Easterly and Levine
(1998) and closely comparable to studies which take as the dependent variable the
growth average over periods longer than a decade, such as Sachs and Warner
(1995). The disadvantage with such a formulation is that the short-term dynamics
of the growth process are not analysed; the advantage is that by including only
structural variables characterising the economy at the start of the decade the prob-
lem of endogeneity is reduced. In the main results presented below OLS was used as
in Easterly and Levine, and Sachs and Warner. The results are then tested against
xed and random effects models.
Civil war is introduced into the regression through three variables which
between them capture the effects both during the war and in the rst ve years
of subsequent peace. The rst, Warmonths, is the months of warfare during the
decade. It measures the concurrent effects of war. The other two focus on the rst
ve post-war years. One variable (Postwar) is the number of these potential recov-
ery months during the decade. The other (Legacy) interacts Postwar with the total
length (in months) of the preceding war. Thus, a war which began in December
1957 and ended in December 1967 would potentially inuence the average growth
rate during the 1960s through there being 84 months of war, through there being
24 months of post-war recovery, and through these post-war months starting from
a legacy of 120 months of war. It would also potentially inuence the average
growth rate during the 1970s through there being 36 months of post-war recovery,
and through these post-war months starting from a legacy of 120 months of war. If
there is a peace dividend irrespective of the duration of the war, then Postwar will
be positive and signicant whereas Legacy will be insignicant. If, however, as
postulated above, the duration of the war determines whether there is a peace
dividend or a war overhang, then Postwar will be negative and signicant and
Legacy will be positive and signicant. There will then be some critical duration
of warfare below which peace yields a war overhang instead of a dividend.
The other explanatory variables are all now fairly standard in growth regressions.
Dummies for all three decades were included in lieu of a constant term, together
with continent dummies. Two endowments at the start of the decade were
included: the level of secondary schooling (entered as a log), and the level of per
capita income. It was found that the best functional form for income was simply its
square. Two country-specic characteristics were included: the degree of ethno-
linguistic fractionalisation and whether the country was landlocked. Easterly and
Levine (1998) found that an index of ethno-linguistic fragmentation, measured as
of 1960, signicantly reduced growth. The index shows the chances of two
randomly drawn people being from different ethno-linguistic groups. Although
potentially the index might be correlated with civil war, Collier and Hoefer
174 economic consequences of civil war
(1998) show that highly fragmented societies have no greater risk of civil war than
homogeneous ones. Sachs and Warner (1995) pioneered the use of a land-locked
location as an explanation of slow growth. Being landlocked is a non-policy im-
pediment to trade and most analysts nd that trade promotes growth.
4
Policy variables were omitted because some of the effects of civil war work
through policies. For example, the portfolio substitution consequent upon civil
war can be expected to increase the premium on the parallel exchange rate.
There is indeed a high correlation between civil war and the size of the premium.
Since the premium is often found to be the most important policy inuence on
growth (Easterly and Levine, 1998; Sachs and Warner, 1995) its inclusion would
detract from the true effect of civil war. A similar argument could be made with
respect to the scal decit. Growth regressions tend to be sensitive to specication.
However, this largely arises because many policy variables tend to be highly corre-
lated so that depending upon which are included, quite different policies may
appear to be important for the growth process. By excluding policy variables we
are left with a less sensitive specication.
5
The results, using heteroscedasticity-consistent standard errors, are reported in
Table 1.
Clearly, the non-war variables are not the focus of our attention (although they
are almost all signicant). However, the `convergence effect' has implications for
post-war recovery: the coefcient on income indicates that poor countries grow
faster than richer countries controlling for the other variables. A country with half
the mean income would grow 2.5% more rapidly than a country with mean
income, evaluated at the mean of other variables. This provides some automatic
peace dividend effect: if a country immiserises itself through civil war it will have an
enhanced post-war growth rate by virtue of its poverty. This effect would predict
that the post-war growth rate would be faster the longer the duration of the
previous war. For example, a country which began a 15-year war with mean
income would add 1.7% to its growth rate in the rst post-war year. However,
since income at the beginning of the decade is one of the explanatory variables, this
effect is largely controlled for and so is over-and-above the effects shown by the two
post-war variables.
Now consider the effects of civil war. During civil war the annual growth rate
is reduced by 2.2%.
6
A 15-year civil war would thus reduce per capita GDP
paul collier 175
..........................................................................................................................................................................
4
Possibly, ethno-linguistic fragmentation reduces growth by a similar process of impeding trade. Gravity
models of trade ows nd that if two countries share a common language trade ows are signicant
increased (Foroutan and Pritchett, 1993). Hence, it may be the linguistic rather than the ethnic aspect of
fragmentation which is costly for growth.
5
For example, if the country or time dummies are dropped, although the performance of the regression
deteriorates the coefcients on the remaining variables are little changed.
6
Since the duration of the war is measured in months and the dependent variable is the average growth
rate during the decade, the coefcient on Warmonths must be multiplied by 120 to arrive at the effect on
the annual growth rate.
by around 30%.
7
The ve post-war years must be assessed through the joint
effects of the Postwar and Legacy variables. The Postwar coefcient is negative
whereas the Legacy coefcient is positive, both being signicant: short wars
cause continued post-war decline, while sufciently long wars give rise to a
phase of rapid growth. During the ve years following a war which only lasted
one year, the growth rate would be 2.1% lower than had the war not happened.
This is not signicantly different from the 2.2% growth cost of the war phase.
By contrast, after a 15-year war the post-war growth rate is enhanced by 5.9%
per annum.
These results are consistent with the war overhang effect predicted for short civil
wars and with the potential for rapid growth after long civil wars implied by the
repatriation of previous capital ight. The prediction that there would be a con-
tinuing exodus of capital after short wars was based upon the proposition that the
post-war environment is less capital-friendly than the pre-war environment. This is
due partly to the persistence of the diversion effect, and partly to the increase in the
perceived risk of civil war (which Collier et al., 1998, have shown to be a rational
expectation). Such a large persistence effect should depress the expected return on
capital relative to its pre-war level for a given capital stock, giving rise to a war
overhang effect. An alternative interpretation is that the risks of renewed war are
176 economic consequences of civil war
Table 1 Effects of civil war and peace upon the growth of GDP
Variable coefcient t-ratio
dummy 1960s 0.0606 7.00*
dummy 1970s 0.0802 6.44*
dummy 1980s 0.0420 4.39*
Africa 70.0159 73.37*
Latin America 70.0235 78.13*
income squared 70.00056 74.10*
schooling (LN) 0.0270 3.05*
fractionalisation 70.00023 74.41*
landlocked 70.0056 71.54
warmonths 70.00018 74.43*
postwarmonths 70.00022 72.11{
legacy 0.0000040 2.25{
* signicant at the 1% level.
{ signicant at the 5% level.
Adj:r
2
:42, F 18:06, n 259.
BreuschPagan test 32.02 (critical value for 95%18.31).
..........................................................................................................................................................................
7
This estimate should be qualied. Because the dependent variable is the average growth rate over a
decade, the range of the Warmonths variable is bounded at 120. A 15-year civil war would be treated as
(say) a 120-month war in one decade and a 60-month war in the next. The application of the coefcient
on Warmonths to a 15-year civil war is thus an extrapolation out of the sample range.
perceived to be higher after a short war than a long one, since short wars may be
inconclusive. In this respect the desired capital stock may be lower after short wars
than after long ones. However, Collier and Hoefer (1998) test for this and nd
that the duration of war does not have a signicant effect on the probability of
subsequent war. Hence, such a perception would not be a rational expectation.
8
In the above OLS analysis the growth performance in each decade has been
treated as an observation ignoring the information that the same country may
appear three times. This is the model used, for example, by Easterly and Levine
(1998). We now test the OLS model against xed and random effects models.
In order to test against the xed effects model the sample has to be reduced from
93 countries to 78 due to the lack of data covering more than one decade for a few
countries. The OLS regression on the full sample was repeated for the sub-sample.
The results are shown in Table 2. The OLS model is not rejected when tested
against the xed effects model. However, the reduction in the sample size reduces
the signicance of all the war-related variables, although the coefcients are little
altered. Warmonths remains signicant at 1% but Postwarmonths and Legacy
become insignicant. Hence, the panel can only be used to test whether
Warmonths remains signicant. Warmonths remains signicant in the xed effects
model.
The random effects model can be tested on the full sample. Again the OLS model
is not rejected when tested against the random effects model. The Warmonths and
paul collier 177
Table 2 OLS versus xed and random effects models
reduced sample full sample
OLS model xed effects random effects
................................................................................................................................................................
Variable coeff. t-ratio coeff. t-ratio coeff. z-ratio
constant 0.0354 3.86* 7 7 0.0416 79.17*
income 7 7 7 7 0.0318 15.72*
income squared 70.00047 72.59{ 70.00045 1.51 70.00274 710.66*
schooling (LN) 0.0309 2.74* 0.0337 1.55 0.0136 1.74{
fragmentation 70.00023 73.75* 70.00023 72.07{ 70.00020 73.38*
landlocked 70.0078 71.72 70.0060 70.70 70.0085 72.26{
warmonths 70.00021 72.96* 70.00020 72.34{ 70.00020 73.83*
postwarmonths 70.00023 71.05 70.00035 71.14 70.00028 71.66{
legacy 0.0000032 0.93 0.0000039 0.88 0.0000042 1.38
* signicant at the 1% level.
{ signicant at the 5% level.
{ signicant at the 10% level.
Fixed effects model versus OLS: F 1.085.
Random effects model versus OLS: Lagrange Multiplier Test: 4.61.
..........................................................................................................................................................................
8
I am indebted to a referee for this point.
Postwarmonths variables remain signicant, although Legacy drops a little below
signicance at the 10% level. Overall, the tests do not provide a basis for rejecting
the OLS results reported in Table 1.
4. The effects of civil war on the composition of GDP
The previous sections have focused upon GDP in aggregate. However, it would be
surprising if, given the large aggregate effects which are caused by civil war, these
were neutral between sectors. Sectors may differ both with respect to their suscept-
ibility to the disruption and diversion effects and because they have different
intensities as between endogenous and exogenous factors. These compositional
effects are potentially testable but unfortunately require much more information
than is presently available in the internationally comparable data sets, especially
since data is particularly weak for civil war economies. The approach taken here is
rst to classify sectors according to their war vulnerability, and then to test the
predicted change in the shares of groups of sectors using an unusually reliable set of
National Accounts. The underlying hypothesis is that civil war affects the composi-
tion of activity partly through increasing the cost of transactions and partly
through reducing the supply of N-factors.
The cost of transactions is increased by civil war through the destruction, dis-
ruption, and diversion effects. The costs of transportation increase as infrastructure
and security deteriorate, and the ability to enforce contracts is reduced as the
institutions of civil society are weakened, trust declines, time horizons shorten
due to uncertainty, and opportunism becomes more protable. Turning to
factor intensities, it is reasonable to expect that during civil war the supplies of
physical and human capital will contract relative to land, buildings, and unskilled
labour. Again this changes the composition of the economy in two ways. Those
sectors which are relatively intensive in capital tend to contract relative to the other
sectors. Further, the fall in demand for physical and human capital reduces the
output of the capital-providing sectors. A nal demand effect is on the defence
industry itself. However, most civil wars occur in low-income countries in which
the defence supply industry is small. In Uganda it was negligible and it is not
considered in the subsequent analysis.
To summarise, activities with any of four characteristics tend to contract: those
which are intensive in either capital or transactions, and those which supply either
capital or transactions. It is quite possible for an activity to have more than one of
these characteristics.
Naturally, National Accounts are not disaggregated according to these character-
istics. It is therefore not possible to make a comprehensive ranking of activities
according to their vulnerability to civil war. However, some activities are suf-
ciently obviously distinctive that it is possible to make partial rankings. Sectors are
now classied according to whether their characteristics are such as to make them
manifestly war vulnerable, or manifestly war invulnerable, leaving those sectors
which are insufciently distinctive in a third, unclassied, category.
178 economic consequences of civil war
The most favoured activity would be one which was transactions-extensive,
capital-extensive, and whose output was neither transactions services nor capital.
Among the National Accounts categories these characteristics apply only to arable
subsistence agriculture. Three groups of activities can be identied as disadvan-
taged. The fall in demand for domestic capital goods applies quite specically to the
construction sector. The fall in demand for transactions services will affect trans-
port, distribution, and nance. The main activity intensive in both capital and
transactions is manufacturing.
This partial ranking assigns activities to three groups: war-invulnerable (arable
subsistence agriculture); war-vulnerable (construction, transport, distribution,
nance, manufacturing); and unclassied (all other activities). Six testable predic-
tions follow from this assignment. The rst three are that during civil war output of
the rst group would expand relative to GDP, output of the second group would
contract, and, if the unclassied group genuinely consists of activities which are not
atypically affected one way or the other, its share of GDP should not be altered. The
remaining three concern the post-war period. The effects of civil war on the
composition of GDP are not automatically reversed in the post-war period because
the effect on factor supplies is not automatically reversed. As discussed in Sections 2
and 3, the period following short civil wars tends to be characterised by continuing
loss of capital. Only after long civil wars is the decline in the capital stock reversed.
Hence, it is only after long civil wars that peace unambiguously reverses the
compositional changes. For such wars the predictions are therefore that the war-
invulnerable group should contract, the war-vulnerable group expand, and the
unclassied activities remain unaltered as shares of GDP.
Evidence on the effect of prolonged civil war on the composition of GDP is
much more problematic than that on aggregate GDP. There is no international
data set on the composition of GDP at constant prices comparable to those
available for aggregate GDP.
9
In place of multi-country panels resort is made to
data on a single country, Uganda, which experienced a 15-year period of
intermittent civil war. Uganda is particularly well-suited for such an analysis
because, by the modest standards of a country which has experienced a prolonged
civil war, its National Accounts are of good quality. Social disturbance on a massive
scale began abruptly in 1972 when President Amin declared `economic war' against
the Asian community, triggering an exodus of human and nancial capital, and an
expansion in the army. The subsequent periodic warfare was resolved only in 1986
when the NRA guerrilla forces took Kampala. For a more detailed discussion of the
effects of the war see Collier and Pradhan (1998).
By 1971 the Ugandan National Accounts were well-established so that the
pre-war composition of the economy is known with reasonable condence.
During the ensuing 15 years the quality of the National Accounts naturally
paul collier 179
..........................................................................................................................................................................
9
The Penn World Tables, which is the most nearly appropriate, does not provide a breakdown of
activities suitable for the breakdown derived in Section 4.
deteriorated. However, the subsequent period of peace has been sufciently long
that they are now of quite high quality. Moreover, the National Accounts have
recently been rebuilt back to 1986. One problem inevitable with long National
Accounts time series is that there is a change of the base year in which the constant
price series are measured. The treatment of this problem is discussed in the notes
to Table 3.
Table 3 reports the composition of GDP according to the three groups of
war-invulnerable, war-vulnerable, and war-neutral activities proposed above. It
compares the structure of the economy at constant 1991 prices in 1971 (the last
pre-disturbance period), 1986 (the end of the civil war), and 19934.
No attempt is made to specify a counterfactual. Qualitatively, it is evident that
the activity proxying war-invulnerability, arable subsistence agriculture, would
normally decline as a share of GDP during the growth process, while those
proxying war-vulnerability would increase, though not necessarily more rapidly
than the war-neutral activities. Fortunately, in the case of Uganda the com-
positional changes are so dramatic that it is reasonable to attribute a substantial
part of them to the effects of civil war.
During the period of the war, the war-invulnerable activity nearly doubled as a
share of GDP whereas in the absence of war it would have been expected to
contract. Conversely, the share of the war-vulnerable group of activities nearly
halved. During the post-war period both of these changes were partially reversed,
although even by 19934 the composition of GDP more closely resembled that at
180 economic consequences of civil war
Table 3 The composition of Ugandan GDP by war-vulnerability (% share of GDP
at 1991 constant prices)
1971 1986 19934
War-vulnerable sectors 42.5 24.0 28.7
high transaction and asset intensity 8.8 4.4 6.0
transaction-providing 21.2 16.1 17.2
asset-providing 12.5 3.5 5.5
War-invulnerable sector 20.5 36.0 32.1
low transaction and asset intensity
supplying neither transactions nor assets
Unassigned activities 37.0 40.0 39.2
Note: The National Accounts provide data at 1966 prices for 1963785 and at 1991 prices from 198293/
4. 1982 was selected as the year to be used for conversion from 1966 to 1991 prices. Since output changes
197182 are only measured at 1966 prices, the conversion of 1971 output to 1991 prices is only
approximate. Sector I in 1971 at 1991 prices is approximated as: [(sector I in 1971 at 1966 prices)/
(sector I in 1982 at 1966 prices].[sector I in 1982 at 1991 prices]. This has the advantage that since 1982
91 GDP was calculated upon a consistent set of denitions of sectors, changes in denitions between the
1966 series and the 1991 series only lead to a mis-estimate to the extent that they alter the growth rate of
the sector between 197182. Total GDP in 1971 at 1991 prices was then calculated as the sum of the
sectoral outputs so revalued. Note that this will differ from a direct adjustment of total GDP in 1971 by
the factor [(total GDP in 1982 at 1991 prices)/(total GDP in 1982 at 1966 prices)]. Sector shares in 1971
at 1991 prices are then sector output/GDP.
the end of the war than that of 1971. Despite these substantial changes in both
directions on the part of the war vulnerable and invulnerable groups, the unas-
signed group maintained a share in the narrow range 3740%, suggesting that it
indeed consists of activities which are at least in aggregate war-neutral.
The six predictions concerning the composition of activity during and after a
prolonged civil war are thus supported on the Ugandan data. This falls well short of
a rigorous test, but it suggests that the hypotheses are worth pursuing on similar
data sets for other countries which have experienced prolonged civil war. With
sufcient data on sector shares at constant prices the war and post-war periods
could be introduced as dummy variables in a time series analysis along similar lines
to that of Section 3, with the dependent variable changed from the growth rate to
sector shares.
5. Conclusion
Civil war is a sufciently devastating phenomenon that it is likely to have large
effects on both the level and composition of economic activity. This paper has
proposed simple theoretical frameworks for these effects and tested them for
war and post-war periods. During civil wars GDP per capita declines at an
annual rate of 2.2% relative to its counterfactual. The explanation proposed in
this paper is that the decline is partly because war directly reduces production
and partly because it causes a gradual loss of the capital stock due to destruction,
dissaving, and the substitution of portfolios abroad. These affect sectors
differentially. The sector intensive in capital and transactions (manufacturing),
and the sectors which supply capital (construction) and transactions (transport,
distribution, and nance), contract more rapidly than GDP as a whole. The sector
with opposite characteristics (arable subsistence agriculture) expands relative
to GDP.
Despite these severe effects of civil war the restoration of peace does not
necessarily produce a dividend. Peace does not recreate either the scal or the
risk characteristics of the pre-war economy: there is a higher burden of military
expenditure and a greater risk of renewed war. The desired capital stock is
consequently lower than had there been no war, although being higher than
that desired during the war. Because downward adjustment of the capital stock
is a slow process, that prevailing at the end of the war may still be above not only
its desired wartime level but its desired post-war level. In this case, which is
inevitable if the war is very brief, the decline in the capital stock can be expected
to continue, yielding a war overhang effect. Empirically, if a civil war lasts only
a year, it was found to cause a loss of growth during the rst ve years of peace
of 2.1% per annum, a loss not signicantly different from that had the war
continued.
However, if the war has been sufciently long the capital stock will have
adjusted to a level below that desired in post-war conditions. In this case capital
repatriation enables the economy to grow more rapidly than during the pre-war
paul collier 181
period. Empirically, this peace dividend for the ending of prolonged civil wars was
found to be large.
Peace also reverses the compositional changes caused by prolonged civil war.
An implication is that after the end of long wars the war-vulnerable activities
experience very rapid growth: the generalised peace dividend is augmented by
compositional change.
Acknowledgements
I would like to thank Anke Hoefer for research assistance and two referees for comments on
a previous draft.
References
Barro, R.J. (1991). `Economic Growth in a Cross Section of Countries', Quarterly Journal of
Economics, 106, 40744.
Bruck, T. (1996). `The Economic Effects of War', M.Phil. thesis, Oxford University.
Colletta, N.J., Kostner, M. and Wiederhofer, I. (1996). The Transition from War to Peace in
Sub-Saharan Africa, World Bank, Washington, DC.
Collier, P. (1994). `Demobilisation and Insecurity: a Study in the Economics of the
Transition from War to Peace', Journal of International Development, 3, 34351.
Collier, P. and Gunning, J.W. (1999). `Explaining African Economic Performance',
Journal of Economic Literature, forthcoming.
Collier, P. and Hoefer, A. (1998). `On the Economic Causes of Civil Wars', Oxford
Economic Papers, 50, 56373.
Collier, P., Hoefer, A. and Soderbom, M. (1998). `On the Duration of Civil War and
Post-war Peace', mimeo, Centre for the Study of African Economics, Oxford University.
Collier, P. and Pradhan, S. (1998). `Economic Aspects of the Ugandan Transition to Peace',
in H.B. Hansen and M. Twaddle (eds), Developing Uganda, James Currey, London.
Dixit, A.K. and Pindyck, R.S. (1994). Investment under Uncertainty, Princeton University
Press, Princeton, NJ.
Easterly, W. and Levine, P. (1998). `Africa's Growth Tragedy: Policy and Ethnic Divisions',
Quarterly Journal of Economics, CXII, 120350.
Foroutan, F. and Pritchett, L. (1993). `Intra-sub-Saharan African Trade: is it too Little?',
Journal of African Economies, 2(1), 74105.
Herbst, J. (1990). `War and the State in Africa', International Security, 14(4), 11739.
Isham, J., Kaufmann, D. and Pritchett, L. (1996). `Government and the Returns on
Investment', Working Paper No. 1550, Policy Research Department, World Bank,
Washington, DC.
Kapuscinski, R. (1988). Another Day of Life, Penguin Books, London.
Knight, M., Loayza, N. and Villanueva, D. (1996). `The Peace Dividend: Military Spending
Cuts and Economic Growth', IMF Staff Papers, 43, 137.
de Melo M., Denizer, C. and Gelb, A. (1996). `From Plan to Market', Policy Research
Working Paper 1564, World Bank, Washington, DC.
182 economic consequences of civil war
Sachs J.D. and Warner, A. (1995). `Economic Reform and the Process of Global Integration',
Brookings Papers in Economic Activity, 25th Anniversary Issue, 1118.
Small, M. and Singer, J.D. (1982). Resort to Arms: International and Civil War, 18161980,
Sage, Beverly Hills, CA.
Small, M. and Singer, J.D. (1994). `Correlates of War Project: International and Civil
War Data: 18161992', Inter-University Consortium for Political and Social Research,
Ann Arbor, MI.
paul collier 183